Quality control of gridded national radar reflectivity data

Archived radar reflectivity data from the WSR-88D network were subjected to three serial procedures to improve their quality for application in automated short-range thunderstorm prediction. The reflectivity data, which are on a 10-km grid that spans the contiguous United States (CONUS), come from two sources. One source is Weather Science Incorporated (WSI), where CONUS reflectivity grids at 2-km resolution are produced and graphically displayed in real time. The other source is the National Weather Service, where real-time, national 10-km grids based on Radar Coded Messages (RCMs) are produced and can be used for applications. The WSI reflectivities are in 16 categories, and they are available four times per hour. The RCM reflectivities are in seven categories, and they are available twice an hour. The quality control (QC) procedures were developed because statistical analyses of the reflectivities and examinations of many plotted reflectivity maps indicated the presence of sporadic non-precipitation echoes and unreliable reflectivities in some geographical locations. The quality control procedures were applied separately for the warm (April - September) and the cool season (October - March).

The first serial procedure is a screen that discarded reflectivities of doubtful validity, as they were from gridboxes beyond 230 km to the nearest operating radar or from gridboxes with substantial terrain occultation. Even though this screen involved significant fractions of the WSI and RCM gridboxes (14.5% and 3.0% , respectively), the volume of discarded reflectivities was quite small (less than 1%), as reflectivities for these gridboxes had been saved only when non-zero categories were reported. The smaller fraction of RCM gridboxes arose because reflectivities for extended-radar-range gridboxes had been discarded previously.

The second procedure consisted of decision tree checks and adjustments of upper (³ 50 dBZ) reflectivities. The restriction to the upper reflectivities is due to the application of cloud-to-ground (CTG) lightning strike data as ground truth for (most) precipitation echoes. The checks involved spatial and temporal continuity in the reflectivities, and consistency of the reflectivities with concurrent and antecedent CTG lightning strikes. Reflectivities identified as invalid were treated by setting them to adjusted values, or, in very rare instances, to missing. For the most part, adjusted reflectivity values were computed from spatial and temporal continuity parameters derived from the radar and lightning grids. Adjustments for rare, very high (³ 65 dBZ) reflectivities in the WSI data were prescribed on the basis of CTG lightning intensity. In general, the checks and adjustments were more elaborate for the WSI data, which was possible because of their increased reflectivity and temporal resolution. Among the ³ 50 dBZ reflectivites, which constituted 0.71% and 0.07% of the full warm and cool season samples, respectively, 5.5% were corrected for the WSI data and 3.8% were corrected for the RCM data. Findings from statistical analyses and case studies indicate a substantial improvement in quality of the upper reflectivities as a result of the corrections. The findings also indicate that most of the suspect echoes resulted from radar beam anomalous propagation.

The third serial procedure was a "catch-all" screen that involved setting reflectivities to missing in geographical locations where seasonal echo relative frequencies (ERFs) were unrealistically high or low, even after applying the first two QC procedures. The anomalously high ERFs appeared as small spurious peaks, which appeared most often in specific mountainous locations. For the ³ 50 dBZ reflectivitives, the spurious peaks were substantially reduced by the decision tree checking procedure. The anomalously low ERFs, which covered larger areas and were not affected by the first two QC procedures, appeared in locations of long radar range, especially within mountainous regions. Radar beam overshooting of precipitation was indicated as the underlying cause. The catch-all screen removed 3.9% (10.7%) of the warm (cool) season reflectivities for the WSI data; the corresponding fractions for the RCM data were 2.3% (10.8%). The larger fractions for the cool season result from the increased incidence of radar beam overshooting, as shallow stratiform precipitation predominates.